What is training-serving skew, and how do you prevent it? Describe strategies for ensuring consistency between training and production environments.

Sample answer preview

Training-serving skew occurs when the data a model sees during inference differs from what it learned during training, causing degraded real-world performance even when offline metrics look good.